A compressive sensing framework for electromyogram and electroencephalogram

Compressive sensing is an emerging technique for reducing data sampling for further processing or transmission. In this paper, we propose an architecture of compressive sensing for electroencephalogram (EEG) and electromyogram (EMG) signals in the telemedicine sensor network. In order to save the hardware cost in encoding-end, the sensing matrix must be simple. Moreover, the decoding algorithm is required with the medium computational complexity under the trade-off between the reconstructed error and the speed of convergence. Accordingly, we propose a modified compressive sensing matching pursuit (MCoSaMP) and the multiple domains decoding method to enhance the performance. The proposed architecture is composed of Bernoulli matrix in encoding-end, Daubechies-4 (DB-4) for EMG signals (DCT for EEG signals), and MCoSaMP algorithm with multiple domains decoding method in decoding-end. The proposed architecture for EEG signals can reduce the percentage root mean square difference (PRD) by 17% compared to other papers. We can achieve the compression ratio (CR) for EEG signals at 0.4 with PRD 9.1%. Moreover, the compression ratio for EMG signals can be achieved at 0.4 with PRD 21.3%. The proposed MCoSaMP with multiple domains decoding method can achieve almost the same PRD with convex optimization for EMG signals. And the complexity can be reduced from O(N³.⁵) to O(m³/(log N)²), where m and N are the number of measurement and length of signal, respectively. Although the PRD of proposed architecture for EMG signals is 6% larger than traditional EMG compression method, the complexity of proposed method in encoding-end is much lower. That achieves the goal of low complexity in encoding-end at telemedicine sensor network.